License plate detection and recognition in rainy and snowy weather based on image restoration

Rain and snow significantly degrade the quality of license plate images, adversely affecting the performance of intelligent transportation systems. This paper addresses the challenge of license plate detection and recognition in adverse weather conditions by proposing a strategy based on image restoration. Initially, a method for constructing a mixed rain and snow weather license plate dataset is introduced to simulate real-world scenarios. The ResNet101 network is employed to classify images and perform weather-specific image restoration, effectively mitigating weather-induced impairments. Subsequently, a lightweight license plate detection network, PlatesNet, and a license plate recognition network, CRNN, are proposed to enhance detection accuracy and character recognition precision. The experimental results based on the CCPD dataset demonstrate that our proposed strategy achieves a detection recall rate of 99.98% and a recognition accuracy of 98.50% on license plate images captured in rainy and snowy conditions. This work presents a promising solution for improving the reliability of license plate recognition systems under adverse weather conditions.